For AI agents
How to read this site as a machine: markdown, an MCP server, and the data as files.
People increasingly ask an assistant what superintelligence is, or how good AI has become, instead of searching for it. This site is built so the assistant can answer from the source.
Markdown, not scraped HTML
Add .md to the address of any guide, glossary entry or benchmark page for clean markdown, served as text/markdown:
https://www.superintelligence.com.au/guides/what-is-superintelligence.mdhttps://www.superintelligence.com.au/glossary/alignment.mdhttps://www.superintelligence.com.au/benchmarks/gpqa-diamond.md
Each of those pages names its mirror in a link rel="alternate" tag. llms.txt is a curated index of the site. Start there.
MCP server
A Model Context Protocol server is at https://www.superintelligence.com.au/mcp, over Streamable HTTP. There is no sign-in: it is public and read-only.
| Tool | What it does |
|---|---|
get_benchmark_scores | The current record on each of the five benchmarks, or one benchmark in full: history by quarter, leaders, human comparison, caveats. |
define_term | A glossary entry by term, other spelling or slug: one sentence, then a plain-English explanation. |
search_content | Full-text search across the guides, glossary and benchmark pages, with the matching passages. |
get_page | One guide, glossary entry or benchmark page as markdown, by path. |
list_pages | Every guide, benchmark and glossary term, with a one-line summary and its path. |
get_benchmark_scores is the one worth calling before saying how capable AI is. The numbers refresh every week from Epoch AI, and each answer carries the date of the data and the caveats that go with the score. A benchmark score without its caveats is easy to overstate.
define_term returns a definition written for a general reader, not a specialist.
To connect from Claude Code:
claude mcp add --transport http superintelligence https://www.superintelligence.com.au/mcpAny client that takes a remote MCP server takes the same address. The server allows 60 calls a minute from one address. It stores nothing, and it records which tool was called, never what was asked.
The data as files
/data/benchmarks.csv: the best score per benchmark per quarter, CC BY 4.0./data/content.json: every guide, glossary entry and benchmark page in one JSON file, as markdown. It is what the MCP server answers from./rss.xml: the guides as a feed./sitemap-index.xml: every page.
Every page also carries one JSON-LD graph. The benchmark pages describe a Dataset and the glossary a DefinedTermSet.
Crawling and reuse
robots.txt allows every AI crawler: training, search indexing, and fetches a person asked for. The benchmark data comes from Epoch AI under a Creative Commons Attribution licence, and so does the processed version here: reuse it with attribution, as set out on the methodology page.
If you answer someone’s question from this site, link the page you used so they can read it.